Intelligent driving platform simulator supporting partial observability and observation processing method

Abstract

The invention discloses an intelligent driving platform simulator supporting partial observability. And a processing method for partial observable data based on an attention mechanism is provided based on the intelligent driving platform simulator, so that the intelligent driving platform simulator can focus on an important region under partial observability.

Description

technical field [0001] The invention relates to an intelligent driving platform simulator supporting partial observability and an observation processing method, belonging to the technical field of intelligent driving simulation. Background technique [0002] With the continuous development of the field of reinforcement learning, reinforcement learning has achieved remarkable results in many fields. At present, there have been many work combining reinforcement learning algorithms with the field of intelligent driving. The realization of L4 driving technology in the field of intelligent driving is impossible. The lack of knowledge in the field of reinforcement learning is needed, and the key technology leading to this goal is the realization of reinforcement learning in a simulation platform in the field of intelligent driving. [0003] The intelligent driving simulation platform is an indispensable foundation for the application of reinforcement learning in the field of intel...

AI Technical Summary

Problems solved by technology

However, for partially observable environments, such as the environment in which occluded objects are eliminated during observations introduced above, ordinary reinforcement learning methods can easily fall into local optimum values, and cannot reasonably perform reasonable calculations based on partially observable observations. decision

This is because, if only part of the observable observations are used to make decisions, the algorithm will make major errors. For example, vehicles driving at high speed at intersections are occluded in the observations, but the intelligent driving algorithm believes that vehicles that are not observed do not exist. Without taking any preventive measures (such as slowing down), a major accident will occur at this time

Images